Open twittidai opened 7 years ago
Will not be fixed. @weidendo suggests to use mpip.
Just for the record: LAIK communication is much more coarse-grained than e.g. MPI. For such coarse-grained, compound communication (pre-calculation of partitionings, transitions, preparing/transforming/execution of corresponding action sequences, and resource usage for this), I am strongly in support for collecting profiling information, both as summary with LAIK_LOG and via an external interface for interactive/realtime feedback.
Perhaps I misunderstood your PR, because it added timing measurements for single MPI calls in the MPI backend, which is useful to understand the overhead of LAIK on top of MPI. This is a different thing and should be left to external profiling tools such as mpip, scalasca, tau or vampir.
Reopening :-)